Should you route important agent work through it?
Our source build failed, and the test command found no test files. That leaves deployment readiness unresolved. TencentDB Agent Memory is a candidate for a contained evaluation when shared assets and team access rules justify operating its services, but this run supplies no evidence that the Hub, proxy or connected agents worked together.
Our run was recorded on August 24, 2026, at commit 97f9465. Command output identifies the MemoryCore package as 2.0.0-beta.1. The supplied v2.0.1 release is dated August 25, 2026, after our run. Its release notes describe later changes; they do not establish that our failed checks were fixed or rerun.
What our sandbox recorded
Our sandbox recorded installation at 103 seconds with exit 0, build at 9 seconds with exit 1, and tests at 5 seconds with exit 1. The harness stores each step's success flag, exit code, elapsed time and timeout status alongside retained output tails. Those are command outcomes, not evidence that an agent workload completed.
The build produced plugin output before failing during script compilation. TypeScript reported that scripts/seed-v2/tsconfig.json did not exist. An earlier stage's completion therefore did not make the overall build successful. The retained error identifies the missing path but does not establish whether the intended repair is to restore that file or change the build script.
Vitest reported "No test files found, exiting with code 1". This was failed test discovery, not a failed assertion or a skipped test. Stored test metrics are null. We have no passing-test total to report, and the stored parser summary does not establish exhaustive unique test coverage.
The install step's successful status coexists with earlier npm errors in its retained output. The stored dependency count also differs from the raw added-package footer. We retain the stored measurements and leave that discrepancy unresolved; the excerpt does not establish the complete installation sequence.
Npm audit recorded 24 dependency advisories: 2 high and 22 moderate. These are recorded advisories, not confirmed exploitable defects or a security assessment. Scan coverage is unknown because the measurement has no scannedPackages field. An installation footer does not establish that missing coverage.
We did not test memory extraction, retrieval quality, permission enforcement, session binding, deletion, restarts, upgrades, private-repository indexing or a running deployment. No new execution accompanies this correction. The MrKeyoor testing methodology explains command recording and sandbox limits; its current text does not establish which procedure version governed this August run.
What the shared hub is intended to solve
TencentDB Agent Memory documents Chat Memory, Skills, Wiki and CodeGraph as shared assets with ownership, visibility and agent bindings. For a team, the decision is whether managing who receives which knowledge warrants adding a hub and proxy to the workflow. These are documented capabilities; our sandbox did not verify asset isolation or correct delivery.
The README describes private, team and restricted visibility, with restricted access through user, role or agent rules. It also describes manual asset binding while fully automated memory routing remains under iteration. CodeGraph prioritizes public HTTPS repositories; private-repository and SSH support is still being refined. Test those limits against your actual requirements before importing project data.
What operating it requires
TencentDB Agent Memory's documented deployment starts Memory Core, Memory Hub and a proxy. Operators configure separate memory and proxy LLM parameters, then point supported agents at the proxy through their model base URL. The proxy therefore becomes another service to configure and verify during an evaluation.
Use synthetic assets to check that an allowed agent can retrieve the intended material and another cannot. Repeat after changing bindings, resetting a session and restarting services. These checks examine the documented sharing model more directly than receiving a successful chat response. They remain proposed checks, not results from our sandbox.
What the OpenClaw report establishes
Issue 1095 reports that the published 1.0.1 package modified an existing OpenClaw installation during postinstall. Our installation output shows an OpenClaw patch script invocation, but it does not show that OpenClaw existed or that any external files changed in our sandbox.
A later comment describes PR 1143 as making the patch opt-in. The supplied thread does not establish that the proposal reached the exact package you intend to install. Inspect that package's install scripts before evaluation. The comment's reported passing tests belong to its author's validation and do not replace our failed checks.
Which alternative fits your architecture?
Mem0 documents a memory layer with SDKs and a fully managed service option. Its README also distinguishes library, self-hosted server and cloud deployment. Consider it when you want application-level memory integration and deployment choices. That is a different starting point from evaluating TencentDB's shared asset hub; we did not compare their retrieval quality or reliability.
Graphiti is a framework for building and querying temporal context graphs, with the surrounding service left to you. Its README assigns the graph database, LLM, embedder, ingestion, tenancy, access control and operations to the developer. Consider it when changing facts and historical relationships drive your design and you intend to build that service yourself.
These are architectural alternatives, not equivalent replacements for the Hub. The supplied evidence does not establish either as a supported complement to TencentDB Agent Memory.
What the retained ratings mean
The editorial scale runs from 1 to 5: 1 means severe obstacles, 2 means substantial unresolved obstacles, 3 means mixed evidence, 4 means substantial supporting documentation or activity with limits, and 5 requires strong evidence across the assessed dimension. Ratings are judgments about the supplied evidence, not performance measurements or production certification.
Setup ease remains 2 because installation completed while build and test discovery failed. Documentation remains 4 for the supplied README's deployment, client, asset and access explanations plus its specialist-guide links; those guides were not executed end to end. Community remains 4 on the narrow evidence of maintainer participation and a concrete follow-up proposal in the supplied issue thread, without measuring adoption or typical response times. Maturity remains 2 because this checkout produced neither a complete successful build nor behavioral test results.